Day 3: Virtualisation, and a Detour into AI Threat Modelling

23-07-2026

What I Did

Finished Virtualisation Basics, the last room I still had left in Module 2 before Cloud Computing Fundamentals. I also went off the standard roadmap for a bit and completed an extra room: AI Threat Modelling Assessment.

What I Learned

The biggest shift from Virtualisation Basics was realizing it isn't really about "running another OS inside your OS." It's about not wasting hardware. Running one application per physical server is like building a separate house for every guest who visits — most of each house sits empty most of the time. Virtualization is building one apartment building instead, with a hypervisor as the building manager dividing it into separate, secure units.

Once that clicked, the rest of the hierarchy fell into place: the hypervisor manages the hardware and hands out virtual machines, each VM is a complete computer with its own OS, and containers go a step lighter by sharing the host's kernel to run just one application instead of a whole separate system. Seeing an actual host monitoring dashboard tied this to real sysadmin work — it's not just about creating VMs, it's about checking whether a given physical server actually has the CPU, memory, and storage headroom for another one before you create it.

AI Threat Modelling Assessment ended up connecting to the same underlying idea, just applied somewhere completely different. The big takeaway there was that AI security isn't about memorizing attack names — it's about identifying which specific component of the AI pipeline (the prompt, the LLM, the retrieval system, the database) is actually exposed, and placing the defense at that exact layer. A Prompt Injection attack targets the prompt and the LLM directly. Sensitive Data Leakage can involve the retrieval system, the database, and the LLM. Data Poisoning happens further upstream, corrupting the database and retrieval system before poisoned data ever reaches the model at all.

Problems and How I Solved Them

In Virtualisation Basics, the one thing that needed a second read was the difference between a VM and a container — both isolate applications, so at first they looked like they solved the same problem. The architecture diagram made it click: a VM includes a full separate operating system, while a container shares the host's existing kernel, which is what makes it lighter but less flexible.

The AI Threat Modelling room had a similar kind of confusion, just with attack names instead of infrastructure terms. Prompt Injection versus Sensitive Information Disclosure looked interchangeable at first, and so did Data Poisoning versus Feature Manipulation. What actually resolved both pairs was asking a single question: where in the AI lifecycle does this attack actually happen — training, inference, or retrieval? Once I started sorting by that instead of by how the attack sounded, picking the right answer got a lot easier.

Visuals

The virtualization hierarchy, from physical hardware down to an individual containerized app:

The AI pipeline, and where each major attack actually lands:

What's Next

Cloud Computing Fundamentals to actually close out Module 2, then on to Operating Systems Basics in Module 3.